My idea with human-like-AI

[earlz]

By human-like, I mean it actually acts like a person would...

but yea, The problem I think with making a real AI is that once you made it to where it could "learn" it would take as long as it would take a human, to get it to act like one...

I mean, with a "smart" learning system though. This way, when it does something you can tell it something simple like "yes" or "no" without having to say something like "yes, unless it is 3:00..then no"
something that just can learn

Also, an assistant for this would be something like a slap on the hand for "bad" and giving it food for "good"

Really, this is too big of a project for me to even design, much less develop..

but anyway..
anyone think this is a good idea?

[Brynet-Inc]

Artificial Intelligence programs rarely get anywhere very commendable.. It's simply not possible yet..

I remember an old "DOS" AI program called "Billy" that stored responses for various questions.. But seriously things will never get much farther then that without some serious hardware improvements.

I myself think instead of trying to "create" intelligence, We should simply find ways of copying our own into machines.

We have brains.. With electrically excitable cells called Neurons.. Why wouldn't it be possible?

What are we but very complex Meat-Popsicles? (Fifth Element pun intended..)

[Alboin]

By human-like, I mean it actually acts like a person would...

but yea, The problem I think with making a real AI is that once you made it to where it could "learn" it would take as long as it would take a human, to get it to act like one...

The thing with human like AI, is that we don't know how the brain works. Sure, we know what parts do what, and so on and so forth, but we don't know everything about it.

Writing human AI is just like writing an emulator for the brain. Now imagine, for a minute, that you are working on Open86. You are implementing a virtual floppy drive. Suddenly you need to know how the floppy drive shuts off it's motor. You look to the documentation. It reads:

Code: Select all

At this time, we have no idea how the floppy drive functions...

What do you do? The only option is to emulate every input and every output. It's as if a C compiler did this:

Code: Select all

if(arg1 == 1 && arg2 == 1)
        return 2;
if(arg1 == 1 && arg2 == 2)
       return 3;

In response to this:

Code: Select all

1 + 2

From this we can see that it is therefore impossible to create an emulator for the human brain that actually works.

Also, if one is to emulate the brain, then one is also to emulate it's environment. So, to actually create the working emulator, you would need to emulate the entire universe. Impossible at the least.

Finally, even if we did know how the brain worked, there is simple too much to store in today's computers. For example:

Code: Select all

The human brain is immense and complex. It contains some one hundred billion neurons,[1][2] which are capable of electrical and chemical communication with tens of thousands of other nerve cells.[1][2] Nerve cells in turn rely on some quadrillion (1015) synaptic connections for their communications.

Quadrillion? Impossible by today's standards.

Just my thoughts on the matter. I've actually thought about this a lot.

[earlz]

yea, we have the harddrive space to store such a program and information, but we have yet any good method of searching for what is "relavant" much less the ram to make it fast at scanning...

[Brynet-Inc]

Yeah, Give it a few million years.. and our brains will be the equivalent of todays floppy drive

I myself like thinking about how wildly complex the designs of our minds probably are.. even processing simple information is probably a remarkably complex task..

I still think everything we learn is stored somewhere.. I myself can close my eyes and picture various events in my life vividly while switching between them in seconds..

How unimaginably cool would it be to access your memories via a computer? Review everything learnt in school/college, Perhaps trade skills with friends.. And more importantly.. Locate your lost keys..

This below quote is enough to make you considerably depressed..

The human brain has a huge number of synapses. Each of the 1012 neurons (ie. a trillion neurons) (including glial cells) has on average 7,000 synaptic connections to other neurons. It has been estimated that the brain of a three-year-old child has about 1016 synapses (10 quadrillion). This number declines with age, stabilizing by adulthood. Estimates vary for an adult, ranging from 1015 to 5 x 1015 synapses (1 to 5 quadrillion).

Sad right?

[earlz]

@alboin
(I posted before your post came up..)
Well we can also reverse engineer it right?
also, we don't need to emulate it's whole enviroment, think of how we feel normal in dreams? even though we think pushing shopping carts is fun and jumping into tubes with mario...
But really..We can emulate some of it's enviroment, I mean we think perfectly well in video games...After a lot of practice, we can WIN games while thinking about something else...Trust me, I know this because I use to play games and start thinking about something and a few minutes later realize that I won a level I never have..
But! you have to actually know the game inside and out..(I can only do that with one game now, but I use to could do that with quite a few of them..)

[AndrewAPrice]

You could have an neural network to aid in repairing computers. For example, trash an operating system in a vm, then run your neural network. Somehow, work out how much of the operating system is fixed in percentage. The neural network will use distributed computing (such Google Compute) to run millions on instances at once, and each instance will 30 minutes to do random actions in the virtual machine. After 30 minutes, all of the instances will be replaced by a newer generation, which will inherit the code of the instance with the higher "repair" percentage.

Of course, even with millions of computers constantly taking part in the simulation, it would take thousands of real years before it actually starts doing something useful.. And then that would just be for one particular problem on one particular system.

Maybe with billions of billions of years, my neural network could become perfect, and even solve a solution on how to travel back in time to seek it's master, then I can become rich!!!!

[earlz]

maybe we'll eventually teach AI to program, of course that would cause the AI to either become evil and destroy your computer, or it would commit suicide after trying to interface with the floppy drive(lol)

[mystran]

Nah, floppy drive is perfectly logical when you finally understand how it works. I wrote a tutorial about it, which reminds me I should remember to extend that tutorial a bit.

[ehird]

A floppy drive is logical.
Our brains' use of logic is sufficiently many and tiny, mixed together, that the entropy makes it seem incredibly close to being not logical. The computation power for that amount of entropy doesn't exist.

If a perfect AI was made today, its personality would be very dull and logical.

[axilmar]

By human-like, I mean it actually acts like a person would...

but yea, The problem I think with making a real AI is that once you made it to where it could "learn" it would take as long as it would take a human, to get it to act like one...

I mean, with a "smart" learning system though. This way, when it does something you can tell it something simple like "yes" or "no" without having to say something like "yes, unless it is 3:00..then no"
something that just can learn

Also, an assistant for this would be something like a slap on the hand for "bad" and giving it food for "good"

Really, this is too big of a project for me to even design, much less develop..

but anyway..
anyone think this is a good idea?

The brain does not work with mathematical logic but with pattern matching on a model of the world. The brain has a really simple algorithm:

1) take whatever input is presented into it (video, audio, touch, smell, taste, other signals).
2) by pattern matching, find a place for the input in the brain's model.
3) by pattern matching, find the next possible states of the model.
4) recall a reaction that is most effective for survival according to the next possible states found in step #3.
5) output the reaction.
6) store the whole experience (steps #1 to #5) as an experience itself.

The real interesting things with the brain are:

1) information association: it seems that anything can be connected to anything, depending on experience.
2) parallelism: searches and pattern matching happens in parallel for all the data.
3) step #6 above is consciousness: it's feedback on the whole experience.
4) the process of input/process/output happens with pipelining: while the new input is being fed to the brain, the brain also processes the previous input and outputs the results of the input before that. It is a continuous process.

Even mathematics is the result of pattern matching. For example, we can do addition because we know the pattern 1+1=2. We can write theorems because we know the pattern of logical deduction. We believe in Gods because we can't prove their existence with deduction, but we have to believe in something anyway if we want to survive. We believe that our wife loves us for what we are, but in reality she simply gets what she wants from us.

The reasons we don't have AI are:

1) research is focused on neural networks; it does not seem to recognize:
a) the mechanism of continuous pattern matching,
b) that the brain builds a model of the world.
c) the whole thing has a motive, i.e. the survival of the entity.
2) the human brain is a parallel machine which we can not yet simulate.
3) the brain has 5 senses: sight, hearing, smell, touch, taste. AI does not have those senses, so even if it could learn all the knowledge of the world, it still could not reach the level of intelligence of the human brain.

[Kevin McGuire]

The brain does not work with mathematical logic but with pattern matching on a model of the world. The brain has a really simple algorithm:

Yeah, but the pattern matching has to be preceded by pattern searching. Also the representation of this pattern has to be expressed and stored.

-> pattern <- [a]
[c] -> pattern <- [a]
[d] -> pattern <- [a]

The (pattern) has to be represented in some form. Take for instance the notion of four image where the brightness is changed. The pattern might represent the magnitude difference of the ratio of RGB in one pixel to the next instead of the absolute value so that:

f(z,y,m) (z[y].m / z[y])
|f(a,i,r) - f(b,i,r)| + |f(c,i,r) - f(d,i,r)| + |f(a,i,g) - f(b,i,g)| + |f(c,i,g) - f(d,i,g)| + |f(a,i,b) - f(b,i,b)| + |f(c,i,b) - f(d,i,b)|


Where you could perform a partial pattern matching. Possibly in case where this creature would be in the same location it was during the day but now at night and the pattern matching would help lock the experience from the day to night time.

Of course there would have to exist other more complex pattern matching maybe on a very high conceptual level since I doubt something would end up standing in the exact same spot as night as it was during the day.

But take a OpenGL maze where there is only four direction you can face with writings and symbols written on the walls. You also adjust the light level to mimic day and night. The AI would hopefully be able to search and validate this pattern while facing a certain wall during the day then at night.

[earlz]

The thing is how to store 3-D images and more over, how to compare and recognize similar 3-D images..the inherit problem is that we really have to emulate a 3-D surface in a 2-D screen...

that is only for recognition of images though..technically, you could make a blind AI though to get rid of that problem..(course, then you might have to have touch and hearing though)

[Zacariaz]

The thing is how to store 3-D images and more over, how to compare and recognize similar 3-D images..the inherit problem is that we really have to emulate a 3-D surface in a 2-D screen...

But are we (humans) really viewing the world in 3d? Sure we have two eyes, seing two slightly different 2d pictures, thus giving us a little depth perception, but i wouldnt really call it 3d. Remember that we can easily be foolled by a 2d picture that looks 3d like.

Anyway, i dont think that is the great problem, however it will take a masive amount of computing power to regongnize object and stuff, and that is the real problem. We dont fully understand how the human brain do this and the methodes used today are slow and consuming.

Remember that the 2d man only see half the cirkel and we only see half the sphere. (well it all depends on the distance and size of the circle/sphere, but more or less...)
We actually have a great deal of trouble visualizing 3d objects compared to 2d objects, exactly because we oly see part of them.

Making the AI blind/deafh and paralysed might not be such a bad start. We still need means of comunication, but it doesnt have to do with light or sound, which are complex to interpet.
One drawback would be the learning process, but maybe a connection to the internet could be the solution?

I believe that a nesacery goal in creating AI would be to anabling it to improve upon it own programming.

[Kevin McGuire]

The thing is how to store 3-D images and more over, how to compare and recognize similar 3-D images..the inherit problem is that we really have to emulate a 3-D surface in a 2-D screen...

What you might do is continually collect frames of video, then at the same time you are constantly processing this stream of frames looking for patterns with previous ones which build the relationships. Frames with out a relationship should disappear. The problem might be the complexity of a pattern. So you might bias the system to build from existing patterns to create a complex pattern. So for every instance of one frame and another you are not starting from scratch. This could be where the conceptual things come from. The area above the frames, made out of patterns. Where you can recall certain patterns that represent something conceptual and have references to frames. Possibly refining these patterns and such while you are still continually storing new frames. Patterns might have a emotional properties associated with them somehow to bias the system to doing pattern matching on more important events instead of just trying to keep up with the head of the image stream.

Interesting too I think is having the ability to not treat a entire frame as a pattern but arbitrary parts in different complex geometric shapes for boundaries.. somehow the complex geometric shapes need a root in the patterns.